A Lumped Element Rat Race

Coupler

A t most frequencies distributed element hybrids, such as the rat race coupleq consume
too much valuable circuit real estate, particular& in MMICs. Lumped element designs
great4 reduce size without large performance costs.

T
Samuel J. Parisi he conventional 180 degree, 3 dB transmis-
MITRE Corporation sion line hybrid, often called a “rat-race cou-
pler” or a “ring hybrid” is depicted in Figure
Bedford, Massachusetts 1 together with its electrical characteristics normal-
ized to a 1 Hcrtz center frequency. Due to symme-
try, the return loss behavior of ports 3 and 4 are not
shown. This abbreviated presentational procedure
will be used throughout the paper. At frequencies
below 18 GHz, this hybrid occupies much morc
area than an equivalent lumped element dcsign.
For example, a 70.7 ohm, quarter wavelength trans-
mission line on a 100 micrometers thick GaAs sub-
strate at 8 GHz is 3325 micrometers in length.
When realized in ring form, it occupies approxi-
mately 32 square millimeters in this medium. An
electrically cquivalent design (at the center frc-
quency) in lumped element form occupies only
about 1 square millimeter, an area savings of over
96%, i.e. a reduction in area by a factor of 30!
Nor is the lumped approach only useful to reduce
the occupied area of couplers in MMIC circuits.
The same methods can be employed to shrink cou-
pler sizes in stripline or microstrip, particularly
below 1 GHz, for which frequencies the wave-
length. even in dielectric. can exceed 1 foot.
 - 270” 0.000
I S-24

dB
-5.000

9 0- -1 0.00
0.500 1.000 1SO0
180’ TRANSMISSION LINE HYBRID -
FREQ HZ
2, 70.71 OHMS

Figure 1a. The distributed element “rat race” coupler. Figure 1 b. Coupling versus frequency of the rat race.

180.0

ANG
0.000

-1 80.0
0.500 1.000 1.500 FREQ-HZ
FREQ-Hz

Figure l c . Phase versus frequency of the outputs. Figure Id. Isolation and return loss versus frequency.

The basis of the design is to derive equivalent
“pi” and ”tee” networks for the transmission line
segments of the “rat-race”. From a theoretical ba-
sis, this is accomplished by writing the ABCD ma-
trix for the corresponding parts of the distributed
and lumped element hybrids and equating the cor-
responding respective terms [ 1, pg. 2291.
the center frequency of the design; however, as will
be seen, the lumped equivalent is good enough to
provide the modest bandwidth often needed for
most applications.
The basis of the design is to derive eqiiivulent
and “tee networks for the transmission line
(‘pi’’ I’

segments
~~

The sume metho& can be employed to shrink
coupler sizes in striplirie or microstrip
This procedure then gives relationships that must
be satisfied if the lumped and distributed circuits
are to be equivalent at the design center frequency.
The equivalence, of course, applies perfectly at only
-
The quarter navelength, or 90 degree. transmis-
sion line segments were modeled as low pass “pi”
networks. The result of the analysis is that the im-
pedance of the elements in the low pass network is
numerically equal to the impedance of the quarter
wave line section being replaced. ‘Thus, for exam-
ple, a lumped pi equivalent of a 90 degree, 70.7 ohm

APPLIED MICROWAVE 1989 & 1990 131

characteristic impedance line scction has inductors 0,000
and capacitors whose reactances are also 70.7 ohms
in magnitude at the frequency for which the equiv-
alence applies.
A “pi” equivalent is preferred to a “tee” in an YB
integrated circuit application because it require5 -5.000
fewer inductors for its realization. For a planar lay-
out, the spiral inductors used exhibit relatively high
resistive losses, consequently their use should be
minimized in lumped element designs.
-1 0.00
0.500 1.ooo 1SO0
The equivalence . . applies p e f e c t h at on& the
, FREQ-HZ
center frequency of the desigq Figure 2b. Calculated coupling versus frequency for the
lumped equivalent.

The three quarter wavelength, 270 degree, trans-

mission line section of the distributed hybrid is re-
placed in the lumped design with a high-pass “tee”
network since the high pass “tee” requires fewer 180.0
inductors than would a high- pass “pi” network.
The matrix operations comprising the analysis of
the circuits are shown in the Appendix. Similarly,
for the high pass circuit, the reactances of the ele- ANG
ments are also equal numerically to the characteris- 0.000
tic impedance of the 270 degree line length being
replaced, at the frequency at which the equivalence
applies.
The final circuit configuration is shown schemati-
cally in Figure 2, along with its calculated pcrform-
ance as functions of normalized frequency. -1 80.0
0.500 1.000 1.500
Table I lists the component values required
FREQ-HZ
(reactances of 70.7 ohms) for lumped element hy-
brids at frequencies as widely disparate as 80 MHL
Figure 2c. Calculated phase response for the lumped cir-
cuit.

h 0.500 1,000 1.500

LUMPED ELEMENT 180” HYBRID FREQ-HZ

Figure 2a. The lumped element rat race equivalent circuit. Figure 2d. Calculated isolation for the lumped circuit.

132 APPLIED MICROWAVE 1989 & 1990

 FREQ(MHz) C(pF) L(nH)
.b
80 28.0 140
1
I

t
I

8000 I 0.28 I 1.40

Table I. The L and C values for lumped equivalent rat

races at 80 and 8000 MHz.

and 8 Ghz. Neglecting parasitics, only frequency

scaling of the component values is required to shift
the design center frequency. For example, a
lumped element design at 80 MHz can be built that
occupies no more than 40 square millimeters.

The result, . . is that the impedance of the

elements in the low pass network is numerical&
equal to the impedance of the quarter wave line
section being replaced.
Simplification of the above circuit, i.e. fewer
components, at the expense of narrower bandwidth
can be realized by employing a high-pass ‘‘pi.’ net-
work instead of a “tee” network for the 270 degree
transmission line segment. The resulting shunt in-
ductors of the high-pass “pi” network have the
same reactance, hence they are resonant with the
shunt capacitors of the low-pass “pi” network at the
design center frequency. Therefore they can be re-
moved without affecting center frequency perform-
ance.
The resulting reduced circuit and its perform-
ance was evaluated but performance calculations
revealed that its bandwidth over which perform-
ance was judged satisfactory was too reduced by
this procedure and so the approach was not pur-
sued.
The photograph of a MMIC lumped rat race hy-
brid designed for operation at 7.95 GHz is shown in
Figure 3, the physical realization of the hybrid ge-
ometry of Figure 2. The coplanar waveguide probes
and circuit pads are evident in the photograph. The
hybrid itself occupies an area of only 1 square milli-
meter.
Through the cooperation of the Rome Air Devel-
opment CenteriEE this design this design was in-
cluded on one of their GaAs hybrid circuit ”found-
ry” runs. However, due to the time constraints that
this entailed, it was not possible to optimize the
circuit’s analytical model with respect to the parasi-
tics created by the interconnection of the L and C
elements.
Figure 3. Photograph of a MMIC lumped rat race hybrid
designed for operation at 7.95 GHz.
Specifically, in the initial design of the lumped
hybrid, the lengths of the interconnecting transmis-
sion lines were assumed to have a length less than
that realized in the final circuit. The effective addi-
tional inductance so created shifted the final center
frequency downward from that of the intended de-
- The measured results, along with computed
sign.
results using an analytical model reflecting the ac-
tual parasitics are shown in Figure 4. The realized
performance is adequate for many applications.
spiral inductors . . . exhibit relntiveb high . . .
losses consequent@ their use should be
minimized
However, given the opportunity for further iter-
ations, using a more accurate model for the parasi-
tics, it should be possible both to design directly for
the intended center frequency as well as to achieve
a broader band over which acceptable performance
is realized.
Summa iy
This technique of simulating line lengths with
lumped element high and low pass filter sections
yields electrical performance comparable to that of
the transmission line circuit, but with significantly
smaller area. The physical realization of the 8 GHz
hybrid circuit demonstrates that calculated per-

APPLIED MICROWAVE 1989 & 1990 133

 -1 0

dB
I I I I I I I
-15
7 I I
I
I
!I
I 1
1
I I
I

1
I
5 6 7 8 9
- 522 Measured

FREQUENCY - GHz -
-
511 Measured
-
-
534 Measured

-
511 Calculated
S22 Calculated
534 Calculated
Figure 4a. Coupling of the Lumped Rat Race.

Figure 4c. Return Loss.

360

270 -
--+- (523.524) Cal
(S13-514) Cal
Acknowledgment
I would like to acknowledge the support of Gary
Scalzi and Capt. William Cowan of RADCI’EE for
the manufacture of the hybrid circuit as well as Jim
180 Devinc and Stephanie Liberacki of MITRE for the
RF tests that were performed on the circuit and
assistance in the paper’s preparation.
v)
w
w
U
0 0
W Appendix
n The ABCD matrix for a losslcss transmission line
-90 is
5 6 7 8 9

FREQUENCY - GHz

Figure 4b. Phase Response.

[c” Bgli [
cosBL
jY,sinPL
jzosinBL
COSPL 1
Lossless transmission line ( a= 0 )

formance is attainable when parasitic effects of the

interconnections are taken into account. For j3L = 90°,
This technique is applicable at all frequencies in
cases in which single and multiple quarter wave-
length, distributed line lengths are required but [cA B
D] = [ py0 j‘O0]
would be too large for practical realization. This
occurs in numerous distributed components such as
quarter wave transformers. t h e 90 degree branch This transmission line segment can be modeled
line hybrid and the Wilkinson power divider. as a low-pass “pi” network as shown below

134 APPLIED MICROWAVE 1989 & 1990

 J* L Lquating thc matrix elements of the “tee” net-
work to t h c matrix element5 of the transmission line
segment yiclds the following results

m m B,X,= 1 and B, = Yo

Therefore, X, = Zo

T h e narrowband version cmploys a high-pass

“pi” network instead of a “tee” network for the 271)
-
1 XLBC
-
jBc(2 X,B,) JXL
1 -XLBc 1 degree transmission line segment. This is shown
below along with its ABCD matrix

Equating the matrix elements of thc “pi” network

to the matrix elements of the transmission line seg-
ment yields the following results

Therefore, B, = yo

The ABCD matrix for a 270 degree transmission

line is

References
1 J F White. ,LficrowaLc Yeniicotidicttor En,qirieerin,g, Van
Nobtrmci Reinhold. N m York. 1982.
2 S. J Parisi. Portions o f thib paper h e r e prewnted at the 1984
This transmission line segment can be modeled MTT-S Intcrndtional 5ymposium i n a paper entitled ‘1XD
as a high-pass “tee”. T h e high-pas\ ”tec” and its Degree Lumped Elcment Hcbrid ’‘
ABCD matrix arc shown below

APPLIED MICROWAVE 1989 & 1990 135